Subject: Re: A new partition handling scheme: wedges
To: None <seifert@sequent.com>
From: Ty Sarna <tsarna@endicor.com>
List: tech-kern
Date: 02/05/1998 13:10:16
[I said: "more than one partition with the same name should be
considered a confiuration error"]

In article <199802051713.JAA00465@eng4.sequent.com> you write:
> "I don't think so, Tim."
> 
> Consider the case where you have more than one version of the OS,
> or more than one OS.  You could have multiple root, usr, var
> partitions, and you want them mounted in different places depending
> on which OS/version you booted.

So you name them "netbsd12-root", "netbsd13-root", "freebsd22-root".
Actually, this type of situation is a an example of a case where named
partitions and being able to rely on unique names would be very useful.
You can make sure when you mount /usr that you get the usr for the right
OS.

> Or you are building a disk to be moved to another machine.
> Or a machine died and you have moved disk(s) from that machine
> to another.

Then you include the target machine's name in the partition name when
installing. You have total freedom to name partitions as you see fit,
and total responsibility to make sure that you pick them in such a way
as to avoid any forseeable conflicts.  I'd recomment something following
the pattern "foovax-nb13-root" for general use.

The whole point of names in this context is to be able to ask for
something by name, and get it without ambiguity.  Without that, why have
names at all? If the names are only reminders, you might as well just
put a README in /dev/dsk saying what's what.

I feel very strongly that you should no more be able to have two
partitions named the same than you should be able to have two logins
with the same name, two network interfaces with the same name, etc.
I also put quite a lot of stock in the fact that the only implementation
of named partitions I know of (RDB) wants unique names.